Stop the landing test racing the shell's exit

[sehkone]

Summary

executor::tests::conformance::landing::a_directory_at_the_destination_fails_the_shell_write fails intermittently, and in the harness rather than in what it tests — run 31291172351, 371 passed and 1 failed, on a commit whose earlier runs were green:

panicked at src/executor.rs:2995:22:
write contents: Os { code: 32, kind: BrokenPipe, message: "Broken pipe" }

PUT_FILE_SCRIPT refuses a directory at the destination in its first four lines, before cat > "$tmp", so on that path the shell exits without reading stdin. run_landing_script is meanwhile writing the contents and .expect()ing the result. Whether eight bytes reach the pipe buffer before the read end closes decides the test, and nothing about the behaviour under test decides that.

The helper now takes BrokenPipe as one outcome of a run and lets the exit status and stderr the caller already asserts on deliver the verdict; every other write error still panics. stdin is closed explicitly before wait_with_output, because the paths that do read it sit in cat until EOF.

PUT_FILE_SCRIPT is unchanged. Refusing before staging anything is the right order, and reading stdin to be polite about a write being refused would be the wrong repair.

Test plan

• Reproduced the CI panic locally by sleeping 250ms between the spawn and the write, which loses the race every time: the pre-change helper fails with exactly that BrokenPipe, and the changed one passes
• With that sleep removed, mod landing's 21 tests pass repeatedly — the explicit drop keeps the paths that read stdin from hanging
• cargo fmt -- --check --config group_imports=StdExternalCrate, cargo clippy --all-targets -- -D warnings, cargo clippy --all-targets --features test-support -- -D warnings, cargo test (350 passed) and cargo test --features test-support (350 passed) all pass

Closes #44